OpenAI has launched GPT-5 Codex, a variant of OpenAI’s latest AI model GPT-5, specialized in agentic coding. It comes as a unique offering, providing real-time assistance to programmers with the ability to autonomously perform long tasks. GPT-5 Codex will fix pre-existing bugs before deployment, complete exceptionally complicated refactoring, and integrate with anything from your cloud environment and IDE to command line terminals, GitHub, and ChatGPT directly.
Codex CLI brings these capabilities to your terminal and allows reading, rewriting, and executing code locally. You can request a specific feature, debug, or ask for an explanation without breaking out of the shell, and keep full connectivity with your ChatGPT account. Here, we will evaluate the performance of Codex CLI to see how it handles real programming tasks, specifically web development and data analysis.
What is Codex CLI?
The new GPT-5-based Codex CLI is an open-source command-line tool forked from OpenAI’s original Codex CLI. It now supports multiple AI providers. It enables developers to interact with large language models in the terminal for tasks like code generation, refactoring, script explanation, and many more.
Codex CLI acts as a smart terminal partner for coding. It listens to your requests and writes code. Codex CLI has access to your shell and file system. If you’re looking for detailed walkthroughs, check out the guides on codex CLI from OpenAI:
https://openai.com/index/introducing-upgrades-to-codex
Installation
Before proceeding with the installation, ensure that the system requirements are satisfied.
System Requirements for Codex CLI
- Operating System: macOS 12+, Ubuntu 20.04+, Windows 11 (via WSL2)
- Node.js Version: 22 or newer (LTS recommended). You can install Node.js by following the instructions on its official page: https://nodejs.org/
- npm: Required (comes with Node.js)
- Git: Optional but recommended
- RAM: 4 GB minimum, 8 GB recommended
- Internet: Required
- Authentication: OpenAI API Key
Installation Steps for Codex CLI
1. Check Node.js version:
node --version #Ensure that it is version 22 or higher.2. Install Codex CLI globally, or using pnpm:
pnpm add -g @openai/codex3. After logging in with your GPT-Plus or Pro credentials, you will see the following screen.
4. Add this line to your shell config file (e.g., ~/.bashrc, ~/.zshrc) for persistence.
5. Run Codex CLI:
codex6. Or pass a prompt directly:
codex “explain this codebase to me”
Things to Keep in Mind
- Node.js and npm (Node Package Manager) must be installed and available in your system PATH.
- For Codex CLI, Node.js 22+ is required, while Gemini CLI works with Node.js 18+.
- Both require authentication via API keys or account login for full functionality.
- On Windows, Codex CLI is best run via WSL2 for full compatibility.
- Remember, when you want to use your CLI for a certain project, make sure you have the current working directory set to the folder path where your project exists.
Running our CLI Instances
Let’s proceed to test the codex with GPT-5 on 3 tasks
- Building a game on a local host
- Building a Daily calorie-tracing application
- Building a Sentimental Analysis Model Workflow
If all the installation steps are complete and the system requirements are met, we can start with the tests.
Task 1 – Building a game on a local host
Goal: Build a 2D arcade shooter with enemies and scoring.
Prompt:
“Create a simple 2D space shooter game using JavaScript and HTML5 canvas (or Python Pygame if preferred). The game should display a spaceship controlled by the player with keyboard arrow keys for movement and the spacebar for shooting. Enemies should spawn at the top and move downward at random intervals. Implement collision detection so bullets destroy enemies and increase the score. Add a basic game-over condition if enemies collide with the player’s ship. Use pixel-art style graphics and ensure smooth animations.”
It’s evident that Codex successfully produced an almost fully defined 2D space shooter, being very clear and precise in its output. The implementation included player controls, enemy spawning, collision detection, and scoring, all organized in a way that was easy to follow. Almost every time, it ran in under one second with only minor adjustments, which was a pleasant surprise.
Output:
The very smooth design, pixel-art graphics, and responsive animations provided a polished feel, despite the fact that it was a simple game. The USP here was Codex’s ability to basically translate a natural-language prompt into a functioning interactive prototype. The game-over logic, score tracking, and even text-displaying messages were all functional as well, demonstrating Codex’s abilities in both logic and presentation.
Task 2: Building a Daily calorie-tracing application
Goal: Build an interactive dashboard with charts and user input.
Prompt:
“Develop a fitness tracker dashboard using HTML, CSS, JavaScript, and a charting library (like Chart.js). Allow the user to log daily steps, calories burned, and hours slept. Display this data in real-time on dynamic charts (bar chart for steps, line chart for calories, pie chart for sleep distribution). Add a weekly summary section that calculates averages and highlights the best/worst day. Include a simple local storage mechanism so user data persists across page refreshes. Style the dashboard with a clean, modern UI.”
Output:
GPT-5 Codex provided the most practical and feature-rich implementation. It integrated Chart.js smoothly, with responsive bar, line, and pie charts updating in real time. The weekly summary section calculated averages accurately, highlighting the best and worst days with clarity. Additionally, Codex handled local storage well, ensuring persistence across sessions, and styled the dashboard with a clean, modern interface.
Task 3: Building a Sentimental Analysis Model Workflow
Goal: Analyze text data for insights.
Prompt:
“Perform sentiment analysis on a dataset. Implement this in a Jupyter Notebook. Start by cleaning the text (remove stopwords, punctuation, and apply tokenization). Use a pre-trained sentiment analysis model (e.g., from Hugging Face Transformers or TextBlob). Compare model predictions with actual ratings and calculate accuracy. Visualize the distribution of positive, neutral, and negative sentiments using charts. End with at least three insights about how customer ratings align (or misalign) with the sentiment analysis model.”
Output:
GPT-5 Codex delivered a clear and structured workflow for sentiment analysis. It implemented text cleaning, tokenization, and model integration with minimal friction, producing reliable results overall. The comparison of predictions with actual ratings worked well, and the accuracy calculation was straightforward. Apart from a small error in a single print statement, everything executed smoothly, including saving sentiment distribution charts for insights.
You can find the complete code form here and download the dataset from Kaggle.
We recently compared Codex with Gemini and Claude Code (read the full comparison here). At the time, we found Claude Code to be the best across all tasks. However, the new GPT-5-based Codex brings significant improvements with respect to its previous version. In the next section, we’ll compare the overall performance of the new GPT-5-Codex in these regards.
Overall Analysis
In my earlier blog, while accessing and evaluating Codex CLI, Gemini CLI, and Claude Code, each exhibited differing strengths. Codex was great at producing fast and iterative code. Gemini was more useful with active search results and running commands. Claude produced depth with great documentation and structure. Each tool can easily be used for debugging or intervention. Though they should be checked by a human prior to deployment.
GPT-5 Codex is a remarkable improvement of the Codex line. As much as it retains the speed and clarity of Codex, it now includes agentic autonomy to handle bug fixes, refactoring, visualization, and run workflow with no /little intervention. Unlike Gemini’s use-case framework and Claude’s structure and depth, or speed of navigation, GPT-5 Codex becomes a combination of both ends: producing nice, clean, reliable, and production-ready solutions digitally at an alarming pace as coded, making it the best tool of the three.
Conclusion
GPT-5 Codex with Codex CLI demonstrates how AI can serve as a reliable partner in coding by converting natural language prompts into functional, production-ready code. From engaging interactive games to polished dashboards and sentiment analysis workflows, it produced polished work with only minimal edits needed, demonstrating its versatility in development applications.
In addition to automation, Codex enhances developer productivity by easily integrating into IDEs, terminals, and cloud environments. The ability to assist meaningfully in real-time with debugging, refactoring, and automation makes it both a functional and relevant modern programming tool. While some of its outputs may continue to need minor revisions, Codex exemplifies the future of agentic coding for programmers to elevate their working, high-potential products.
Hello! I'm Vipin, a passionate data science and machine learning enthusiast with a strong foundation in data analysis, machine learning algorithms, and programming. I have hands-on experience in building models, managing messy data, and solving real-world problems. My goal is to apply data-driven insights to create practical solutions that drive results. I'm eager to contribute my skills in a collaborative environment while continuing to learn and grow in the fields of Data Science, Machine Learning, and NLP.
